Land surface VIS/NIR BRDF module for RTTOV-11: Model and Validation against SEVIRI Land SAF Albedo product

This proceeding describes the scientific approach and a preliminary validation of the visible and near infrared land Bidirectional Reflectance Distribution Function (BRDF) module for RTTOV version 11. The module provides global (at a spatial resolution of 0.1°) and monthly means land surface BRDF for any instrument with channel’s central wavelength between 0.4 and 2.5 μm. It is based on a reconstructed hyperspectral BRDF from the seven channels of the operational global MODIS 16-days BRDF kernel-driven product MCD43C1 and a principal component analysis (PCA) regression method applied on the USGS hyperspectral measurements version 6 database for soils and vegetation surfaces. The module adopts a methodology similar with the UWiremis infrared land surface emissivity module developed for RTTOV. A preliminary validation of the BRDF against the SEVIRI Land SAF albedo product show a good correlation of the RTTOV module when applied on three SEVIRI visible and near infrared channels. Introduction In the next version of RTTOV (version 11) there will be a possibility for the users to simulate clearsky satellite observations in the visible and in the near infrared spectral regions. In clear-sky situations over land, the major contribution to the signal simulated at the top of atmosphere comes from the surface. As for thermal or microwave spectral regions, the surface optical properties exhibit strong spectral signatures based on the surface type. Furthermore, in the visible and near infrared spectral regions, the surface optical properties also exhibit a strong geometrical dependency, which depend on the solar and on the satellite directions. To describe the spectral and the geometrical dependences of the surface, the surface optical properties are represented by the Bidirectional Reflectance Distribution Function (BRDF). Additionally, the surface optical properties of vegetation-covered area present a non-negligible seasonal dependency. In order to take into consideration the spectral, geometrical and temporal dependencies of the BRDF in visible and near infrared regions over land to provide a good BRDF estimate for RTTOV, we used a combination of hyperspectral reflectance laboratory measurements to take into account the spectral variability of land surface’s reflectivity, and of global and seasonal satellite-based measurements of the BRDF to take into account both surface type and geometrical and seasonal variabilities of the surface optical properties. The methodology was developed in a similar way as it has been done for the RTTOV UWiremis infrared land surface emissivity module (Borbas and Ruston 2010). Here, the hyperspectral laboratory measurements database from the United States Geological Survey (USGS) has been used as well as the BRDF retrieval product from the MODIS Land team. The MODIS MCD43C1 product provide three parameters that allow the full description of the BRDF at 7 visible and near infrared bands bands (at 0.470 μm, 0.555 μm, 0.659 μm, 0.865 μm, 1.24 μm, 1.64 μm and 2.13 μm). The latter wavelengths of the MODIS bands have been chosen as hinge points for a Principal Component Analysis (PCA) on the hyperspectral laboratory measurements database. Then the MODIS measurements allow to constrain the reconstruction of a BRDF spectra that are used next to interpolate the BRDF of any instrument with central wavelengths in the visible and near infrared. The study is organized as follows. Section 2 present the USGS laboratory spectra selected for visible and near infrared spectral regions, i.e. between 0.4 and 2.5 μm. Section 3 gives a short description of the PCA analysis as well as the comparison between the laboratory measurements and their reconstructed spectra. Section 4 provides the mathematical relation between the MODIS product parameters and the BRDF. Section 5 gives details of the resampling of the original global MODIS 16-days BRDF dataset at 0.05° spatial resolution to a coarser global monthly means at 0.1° spatial resolution, as well as the development of a simplified quality mask for RTTOV. In Section 6, we performed a preliminary evaluation of the RTTOV BRDF module by comparing monthly mean RTTOV Black-Sky Albedo (BSA) values with one day of Land SAF SEVIRI albedo product in the three visible and near infrared SEVIRI bands (at 0.6 μm, 0.8 μm and 1.6 μm). Conclusion and perspectives are given in Section 7. The USGS spectral library The USGS spectral library version 6 is freely available at http://speclab.cr.usgs.gov/spectral-lib.html. It is a compilation of over 1300 hyperspectral reflectance spectra measured in laboratory, field campaigns or aircraft-based instrumentations of natural and man-made materials (Clark et al. 2007) that cover ultraviolet to mid infrared spectral regions (from 0.2 to 150 μm) at different spectral resolution between 0.002 and 0.03 μm. The database covers six classes of surface type (mineral, soil, coating, water, man-made and vegetation). The spectral region of each laboratory spectra differs by surface types. For the purpose of its application to RTTOV for land surface in visible and NIR spectral domains, we selected all spectra that cover the range between 0.4 to 2.5 μm for soil and vegetation surface types. Based on the overall database, the spectral range criterion and some visual inspection of the spectra quality, we selected a total number of spectra of 126 (100 spectra for soils and 26 spectra for vegetation). Fig. 1 shows the spectral variation of the reflectance for soil (Fig. 1a) and vegetation (Fig. 1b) surface types. Fig. 1: The selected USGS spectra for soil (a) and vegetation (b) surface types. (a) (b) The spectral resolution of the selected 126 spectra has been uniformed to 0.01 μm leading to 2101 spectral points for each spectrum. The Principal Component Analysis (PCA) on the selected laboratory spectra The first Principal Components (PCs or eigenvectors) of 126 selected laboratory spectra with wavelength resolution of 0.01 μm, were regressed against 7 hinge points corresponding to the central wavelength of MODIS. Fig. 2 illustrates the 126 laboratory spectra (Fig. 2a) and the reconstructed ones using 6 PCs (Fig. 2b). The number of 6 PCs was found to be most optimal by giving the best agreement between the original laboratory measurements and reconstructed spectra. This number was used hereafter in this study and was coded as default value in the RTTOV BRDF module. Fig. 2: The 126 selected USGS laboratory spectra (a) and their reconstructed spectra using 6 PCs (b). The MODIS BRDF kernel-driven product The MODIS BRDF kernel-driven product (named MCD43C1) is based on a 16 day period of acquisition and is provided globally at 0.05° spatial resolution. This product makes use of both Terra and Aqua satellites. The Collection 5 version is freely available at the following address: https://lpdaac.usgs.gov/products/modis_products_table/brdf_albedo_model_parameters/16_day_l3_0_ 05deg_cmg/mcd43c1. The product contains three BRDF kernel model parameters (fiso, fvol and fgeo) for the full description of the BRDF at 7 MODIS bands as well as quality flags. The BRDF is calculated by using the semi empirical linear model of Ross-Li (Lucht et al. 2000) that is given by: ) , , ( ) ( ) , , ( ) ( ) ( ) , , , ( φ θ θ λ φ θ θ λ λ λ φ θ θ ∆ + ∆ + = ∆ sol sat geo geo sol sat vol vol iso sol sat K f K f f BRDF , (1) where θsol, θsat and ∆φ are the solar zenith angle, the satellite zenith angle and the azimuth difference between satellite and solar directions, respectively. λ is the wavelength. The first BRDF kernel model (a) (b) parameter fiso is due to isotropic scattering. The second BRDF kernel model parameter fvol is due to radiative transfer-type volumetric as from horizontally homogeneous leaf canopies. The third BRDF kernel model parameter fgeo is due to geometric-optical surface scattering as from scenes containing 3D objects that cast shadows and are mutually obscured from view at off-nadir angles. The BRDF model kernel Kvol is given by: 4 cos cos sin cos ) 2 / ( ) , , ( π θ θ π φ θ θ − + Θ + Θ Θ − = ∆ sat sol sat sol vol K , (2) where φ θ θ θ θ ∆ + = Θ cos sin sin cos cos cos sat sol sat sol and Θ is the scattering angle. The BRDF model kernel Kgeo is given by: ) cos 1 ( 2 1 sec sec ) , , ( ) , , ( Θ + + − − ∆ Ο = ∆ sat sol sat sol sat sol geo K θ θ φ θ θ φ θ θ , (3) where ) sec )(sec cos sin ( 1 ) , , ( sat sol sat sol t t t θ θ π φ θ θ + − = ∆ Ο , (4) with sat sol sol sat D t θ θ φ θ θ sec sec ) sin tan (tan 2 cos 2 2 + ∆ + = , (5) and with φ θ θ θ θ ∆ − + = cos tan tan 2 tan tan 2 2 sol sat sol sat D . (6) Resampling the MODIS dataset to get a monthly BRDF and an associated quality index The RTTOV BRDF module provides an atlas of monthly mean BRDF values on a fixed 0.1° spatial resolution. For that, we reassembled the original MODIS MCD43C1 BRDF kernel-driven product based on best pixels selected from a simplified mask that is rested itself on the original MODIS MCD43C1 quality flag. The methodology was done in two steps: (1) a spatial averaging and (2) a temporal averaging. The first step (i.e. the spatial averaging) is applied for each original MODIS MCD43C1 product (i.e., for a 16 days period). We used the original MODIS flags in order to get a simplified flag at a coarser spatial resolution (from original pixel at 0.05° to final pixel at 0.1°, so on a 4 by 4 original pixels basis). The second step (i.e. the temporal averaging) leans on the simplified flag obtained at the first step and on the selection of the best final pixel over the 16-days regridded MODIS data within a month. The two steps are described hereafter. The first step makes use of three original MODIS MCD43C1 quality flags. The first one is the MODIS BRDF quality flag (called hereafter sq) that have 5 different values: sq=0 for best quality retrieval, i.e. for 75% or more with best full inversions, sq=1 for good quality retrieval, i.e. for 75% or more with full inversions, sq=2 for mixed retrieval, i.e. for 75% or less full inversions and 25% or less filled values, sq=3 for all magnitude inversions or 50% or less filled values, sq=4 for 50% or more filled values. The two other MODIS BRDF flags are named as “Percent inputs” (that gives the percentage of inputs between 0 and 100 % and called hereafter sp) and “Percent snow” (that gives the percentage of snow between 0 and 100 % and called hereafter sps). From these three information, we developed a simplified flag on the regridded 16-day product (i.e. on each 4 by 4 original pixel at 0.05° within one final pixel at 0.1°) with different iterative tests. If the first test is validated, then a flag number is associated. If not, the second test is done, and so on...Table 1 describes the different tests and associated flag numbers. For each test, if at least one original pixel is found within the final pixel, the flag number is associated. The final value of the three BRDF kernel model parameters f is calculated by the value of the best original pixel or by the mean value if more than one original pixels validating the test are found. Table 1: Test criterion for the flag number of the RTTOV simplified mask. Criterion Test order Flag number sq sp sps Description 1 1 0-1 ≥ 80% 0% No snow, best and good quality for 80% inputs or more 2 2 0-1 ≥ 80% 100% Snow, best and good quality for 80% inputs or more 3 3 2-3 ≥ 80% 0% No snow, medium quality for 80% inputs or more 4 4 2-3 ≥ 80% 100% Snow, medium quality for 80% inputs or more 5 5 < 4 < 80% ≠0% or ≠100% Remaining pixels, bad quality 6 6 4 0–100% 0–100% Filled values, bad quality In the second step, the flag numbers and the values of three BRDF kernel model parameters f are used to calculated a monthly simplified mask and the monthly means of f by using all original MODIS products within a month. For that, we searched for the best final pixel within a month, following: Mask = 0 for water surface, Mask = 1 if at least one final pixel with flag=1 is found, Mask = 2 if at least one final pixel with flag=3 is found, Mask = 3 if at least one final pixel with flag=6 is found, Mask = 4 if at least one final pixel with flag=2 or simplified flag=4 is found, Mask = 5 if at least one final pixel with flag=5 is found, Mask = 6 if no BRDF data (following the land/sea mask from UWIR emissivity atlases). By this way, we kept only the best information from original MODIS MCD43C1 product into the final RTTOV BRDF atlas. If more than one final pixel are found within a month, we calculated a new mean of the three BRDF kernel model parameters f. Fig. 3 shows the monthly RTTOV BRDF mask for January 2007 (Fig. 3a), April 2007 (Fig. 3b), July 2007 (Fig. 3c) and October 2007 (Fig. 3d). Night persistent areas in high latitude or cloudy persistent areas (like in India in July, see Fig. 3c) are classified as no data. Antarctic and Artic areas, as well as snow-covered areas in winter season are well classified. Areas permanently classified as medium like in the north part of South America, in middle Africa or in Asia are explained by the difficulty to retrieve BRDF in presence of strong aerosols loading or cloud contamination. Fig. 3: RTTOV BRDF atlas mask for 2007: January (a), April (b), July (c) and October (d). Validation of the RTTOV BRDF atlas with the Land SAF Black-Sky Albedo product For any given location (in latitude and longitude), month, geometry and instrument channel’s central wavelength, the RTTOV-11 BRDF module provides the BRDF estimate and the associated quality index that is extracted from the original MODIS quality flags. For the validation of the RTTOV BRDF module, we used SEVIRI Land SAF product for the 25 August 2011 averaged at 0.1° spatial resolution from the SEVIRI original spatial resolution. We were not able to validate directly the BRDF since the BRDF is not an operational product from the Land SAF team. We used the SEVIRI Land SAF Directional Hemispherical Reflectance or Black-Sky Albedo (BSA) product. The Land SAF Land Surface Albedo product is documented at the Land SAF website (http://landsaf.meteo.pt). The RTTOV BSA αbs is calculated from the MODIS BRDF kernel-driven product and is given by the following equation (Lucht et al. 2000): ) )( ( ) )( ( ) )( ( ) , ( 3 2 2 1 0 3 2 2 1 0 3 2 2 1 0 sat geo sat geo geo iso sat vol sat vol vol vol sat iso sat iso iso iso sat bs g g g f g g g f g g g f θ θ λ θ θ λ θ θ λ λ θ α + + + + + + + + = , (7) where the different g coefficients are given in Table 2. (a) (b)